irqchip/gic-v4: Fix occasional VLPI drop

Heyi Guo · Marc Zyngier · commit 6479450f72c1 · 2019-01-28T18:01:10.000Z
1. In current implementation, every VLPI will temporarily be mapped to the first CPU in system (normally CPU0) and then moved to the real scheduled CPU later. 2. So there is a time window and a VLPI may be sent to CPU0 instead of the real scheduled vCPU, in a multi-CPU virtual machine. 3. However, CPU0 may have not been scheduled as a virtual CPU after system boots up, so the value of its GICR_VPROPBASER is unknown at that moment. 4. If the INTID of VLPI is larger than 2^(GICR_VPROPBASER.IDbits+1), while IDbits is also in unknown state, GIC will behave as if the VLPI is out of range and simply drop it, which results in interrupt missing in Guest. As no code will clear GICR_VPROPBASER at runtime, we can safely initialize the IDbits field at boot time for each CPU to get rid of this issue. We also clear Valid bit of GICR_VPENDBASER in case any ancient programming gets left in and causes memory corrupting. A new function its_clear_vpend_valid() is added to reuse the code in its_vpe_deschedule(). Fixes: e643d80 ("irqchip/gic-v3-its: Add VPE scheduling") Signed-off-by: Heyi Guo <guoheyi@huawei.com> Signed-off-by: Heyi Guo <heyi.guo@linaro.org> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
@@ -2059,6 +2059,29 @@ static int __init allocate_lpi_tables(void)
 	return 0;
 }
 
+static u64 its_clear_vpend_valid(void __iomem *vlpi_base)
+{
+	u32 count = 1000000;	/* 1s! */
+	bool clean;
+	u64 val;
+
+	val = gits_read_vpendbaser(vlpi_base + GICR_VPENDBASER);
+	val &= ~GICR_VPENDBASER_Valid;
+	gits_write_vpendbaser(val, vlpi_base + GICR_VPENDBASER);
+
+	do {
+		val = gits_read_vpendbaser(vlpi_base + GICR_VPENDBASER);
+		clean = !(val & GICR_VPENDBASER_Dirty);
+		if (!clean) {
+			count--;
+			cpu_relax();
+			udelay(1);
+		}
+	} while (!clean && count);
+
+	return val;
+}
+
 static void its_cpu_init_lpis(void)
 {
 	void __iomem *rbase = gic_data_rdist_rd_base();
@@ -2144,6 +2167,30 @@ static void its_cpu_init_lpis(void)
 	val |= GICR_CTLR_ENABLE_LPIS;
 	writel_relaxed(val, rbase + GICR_CTLR);
 
+	if (gic_rdists->has_vlpis) {
+		void __iomem *vlpi_base = gic_data_rdist_vlpi_base();
+
+		/*
+		 * It's possible for CPU to receive VLPIs before it is
+		 * sheduled as a vPE, especially for the first CPU, and the
+		 * VLPI with INTID larger than 2^(IDbits+1) will be considered
+		 * as out of range and dropped by GIC.
+		 * So we initialize IDbits to known value to avoid VLPI drop.
+		 */
+		val = (LPI_NRBITS - 1) & GICR_VPROPBASER_IDBITS_MASK;
+		pr_debug("GICv4: CPU%d: Init IDbits to 0x%llx for GICR_VPROPBASER\n",
+			smp_processor_id(), val);
+		gits_write_vpropbaser(val, vlpi_base + GICR_VPROPBASER);
+
+		/*
+		 * Also clear Valid bit of GICR_VPENDBASER, in case some
+		 * ancient programming gets left in and has possibility of
+		 * corrupting memory.
+		 */
+		val = its_clear_vpend_valid(vlpi_base);
+		WARN_ON(val & GICR_VPENDBASER_Dirty);
+	}
+
 	/* Make sure the GIC has seen the above */
 	dsb(sy);
 out:
@@ -2755,26 +2802,11 @@ static void its_vpe_schedule(struct its_vpe *vpe)
 static void its_vpe_deschedule(struct its_vpe *vpe)
 {
 	void __iomem *vlpi_base = gic_data_rdist_vlpi_base();
-	u32 count = 1000000;	/* 1s! */
-	bool clean;
 	u64 val;
 
-	/* We're being scheduled out */
-	val = gits_read_vpendbaser(vlpi_base + GICR_VPENDBASER);
-	val &= ~GICR_VPENDBASER_Valid;
-	gits_write_vpendbaser(val, vlpi_base + GICR_VPENDBASER);
-
-	do {
-		val = gits_read_vpendbaser(vlpi_base + GICR_VPENDBASER);
-		clean = !(val & GICR_VPENDBASER_Dirty);
-		if (!clean) {
-			count--;
-			cpu_relax();
-			udelay(1);
-		}
-	} while (!clean && count);
+	val = its_clear_vpend_valid(vlpi_base);
 
-	if (unlikely(!clean && !count)) {
+	if (unlikely(val & GICR_VPENDBASER_Dirty)) {
 		pr_err_ratelimited("ITS virtual pending table not cleaning\n");
 		vpe->idai = false;
 		vpe->pending_last = true;
